Section: New Results

Users Modeling and Designing Interaction

User-centered Heuristics for the Control of Personal Data

Participants : Alain Giboin, Patrice Pena, Fabien Gandon.

This work (done in collaboration with Karima Boudaoud and Yoann Bertrand, SPARKS, I3S, in the context of the PadDOC FUI project) led to the elaboration and the evaluation of a set of user-centered heuristics and a procedure for designing and evaluating systems allowing the control of personal data. The elaboration of the heuristics was based on: (1) the transposal of Nielsen's heuristics and of Scapin and Bastien's ergonomic criteria to the control of personal data ; (2) the user centering of the Privacy-by-Design notion of integrated privacy; and (3) the integration of Altman's interaction approach to privacy.

Needs Analysis of the Target Users of the WASABI musical search platform

Participants : Alain Giboin, Isabelle Mirbel, Michel Buffa, Elmahdi Korfed.

In the context of the ANR project WASABI, we performed an analysis of the needs of the target users of the future WASABI platform. This analysis has been reported in an internal report.

Modeling the Users of Collaborative Ontology Building Environments

Participant : Alain Giboin.

We undertook a study on the evolution of the user model of collaborative ontology building environments (COBEs). By a user model – or a contributor model – we refer to the representation that COBEs designers have of the users of their systems and more generally of the actors contributing to the building of ontologies. This study aimed at emphasizing the importance to get a better knowledge of potential COBE contributors in order to design collaborative tools better suited to these contributors. The study was published in [55]. In this paper, we describe: (1) the method we used to study the evolution of the user/contributor model; (2) the evolution of the model (in terms of user types, user characterizations, and user’s environment characterizations); (3) the parallel evolutions of: (a) the methods of COBEs design, (b) the systems themselves, and (c) the methods of collaborative ontology building; we mention some evolution perspectives envisioned by the designers.

Design of a User-Centered Evaluation Method for Exploratory Search Systems

Participants : Emilie Palagi, Alain Giboin, Fabien Gandon.

This work was undertaken in the context of the PhD of Emilie Palagi, in cooperation with Raphaël Troncy (EURECOM). Our method takes into account users' exploratory search (ES) behavior and is based on a cognitive model of an ES task. We specially work on Discovery Hub (Wimmics project – Inria) and 3cixty (EURECOM project) ESSs. During the third year of the PhD, we continued the evaluation of our model of exploratory search by comparing it to video records of seven other ES sessions on Discovery Hub, Frankenplace and 3cixty. We analyzed the videos with the same methodology: we wrote down the different chains of the different model's features used by the users in their ES session. For all the records we were able to identify the features of our model and extend our table of observed possible transitions between the model's features. From this analysis, we conclude that our model of ES can express the users' activity during an ES task. This work was partially published in [49].

Based on the ES model's features and the possible transitions between them, we designed two different evaluation and design methods of ES systems which do not necessarily involve users:

• Without users: Heuristics of ES and a procedure to use them. These heuristics are principles for the interaction design. The Heuristics of ES can be used several times along the design process of the ES system (in the design and evaluation phases). We presented the heuristics and evaluated them. This work was published in [48].

• With users: a guide for the elaboration of a customizable test protocol. The goal of the test is to analyze ES session records in order to find the model's features. In this guide, we give indications to customize the protocol and prepare users tests. We focused on two model-based elements of this customizable test protocol: a protocol for the elaboration of exploratory search tasks, and a video analysis grid for the evaluation of recorded exploratory search sessions.

Supporting Learning Communities with Intelligent services

Participants : Oscar Rodríguez Rocha, Catherine Faron Zucker.

The Système Intelligent d'Enseignement en Santé 3.0 (SIDES 3.0), (Intelligent Health Education System 3.0), is a 3 years project funded by the French National Agency for Research (ANR) within the framework of the call for projects DUNE 2016. It builds upon a national Web platform, the Système Informatique D’Evaluation en Santé (SIDES) (Health Assessment Information System), used since 2013 by the faculties of medicine in France which enables them to perform all of their validation exams on tablets, providing them with automatic corrections. It contributes to the preparation of medical students to perform the Epreuves Classantes Nationales (ECN) informatisées (ECNi) (Computerized National Qualifying Events) which have been successfully held in France in June 2016 (8000 candidates simultaneously throughout France). The SIDES platform is administered by the 35 medicine faculties in France and is used by more than 70,000 students throughout their training. The system is also used to prepare students for ECNi. Over the last 3 years, more than 4 million clinical cases (made up of 15 questions each) have been performed by students (all activities combined).

Building on this success, the SIDES 3.0 project aims to upgrade the SIDES solution to an innovative solution providing the user with intelligent learning services based on a modelization of the pedagogical resources with Semantic Web models and technologies. It is coordinated by the Université Numérique Thématique (UNT) en Santé et Sport (http://www.uness.fr). This structure offers an ideal national positioning for support and coordination of training centers (UFR) and also offers long-term financial sustainability. In this framework, we focus on developing and applying adaptive learning approaches to automatic quiz generation from existing questions, and quiz recommendation adapted to user profiles and learning contexts, to allow medical students to better achieve their educational objectives by answering quizzes [50], [51].

Explainable Predictions Using Product Reviews

Participants : Elena Cabrio, Fabien Gandon, Nicholas Halliwell, Freddy Lecue, Serena Villata.

This is a joint work between Accenture and Wimmics team, funded by Accenture. The goal of this project is to design a recommender system that returns explainable predictions to the user, incorporating text from the product reviews in the explanation. To start, we have replicated results from current state of the art methods. We then gathered a dataset of Amazon books and corresponding reviews, and ran the current state of the art algorithm on our dataset. The next steps will be to build a deep learning model to outperform the current state of the art algorithm, and develop a method to explain the predictions to the user using the product reviews.

Argument Mining

Participants : Elena Cabrio, Fabien Gandon, Claude Frasson, Andrea Tettamanzi.

We have published a survey paper about Argument Mining at IJCAI [61]. Argument mining is the research area aiming at extracting natural language arguments and their relations from text, with the final goal of providing machine-processable structured data for computational models of argument. This research topic has started to attract the attention of a small community of researchers around 2014, and it is nowadays counted as one of the most promising research areas in Artificial Intelligence in terms of growing of the community, funded projects, and involvement of companies. In this paper, we presented the argument mining tasks and we discussed the obtained results in the area from a data-driven perspective. An open discussion highlights the main weaknesses suffered by the existing work in the literature and proposes open challenges to be faced in the future.

Together with two colleagues from FBK Trento (Italy), we applied argumentation mining techniques, in particular relation prediction, to study political speeches in monological form, where there is no direct interaction between opponents. We argued that this kind of technique can effectively support researchers in history, social and political sciences, which must deal with an increasing amount of data in digital form and need ways to automatically extract and analyse argumentation patterns. We tested and discussed our approach based on the analysis of documents issued by R. Nixon and J. F. Kennedy during 1960 presidential campaign. We relied on a supervised classifier to predict argument relations (i.e., support and attack), obtaining an accuracy of 0.72 on a dataset of 1,462 argument pairs. The application of argument mining to such data allowed not only to highlight the main points of agreement and disagreement between the candidates' arguments over the campaign issues such as Cuba, disarmament and health-care, but also an in-depth argumentative analysis of the respective viewpoints on these topics. The results of this research have been published at AAAI [58].

In this direction, we have also, in collaboration with the Heron Lab of the University of Montreal, presented an empirical study about the relation between argumentative persuasion and emotions. Argumentative persuasion usually employs one of the three persuasion strategies: Ethos, Pathos or Logos. Several approaches have been proposed to model persuasive agents, however, none of them explored how the choice of a strategy impacts the mental states of the debaters and the argumentation process. We conducted a field experiment with real debaters to assess the impact of the mental engagement and emotions of the participants, as well as of the persuasiveness power of the arguments exchanged during the debate. Our results showed that the Pathos strategy is the most effective in terms of mental engagement. The results of this research have been published at FLAIRS [60].

Together with Souhila Kaci (LIRMM) and Leendert van der Torre (University of Luxembourg), we have proposed a formal framework to reason about preferences in abstract argumentation. Consider an argument A that is attacked by an argument B, while A is preferred to B. Existing approaches will either ignore the attack or reverse it. We introduced a new reduction of preference and attack to defeat, based on the idea that in such a case, instead of ignoring the attack, the preference is ignored. We compared this new reduction with the two existing ones using a principle-based approach for the four Dung semantics. The principle-based or axiomatic approach is a methodology to choose an argumentation semantics for a particular application, and to guide the search for new argumentation semantics. For this analysis, we also introduced a fourth reduction, and a semantics for preference-based argumentation based on extension selection. Our classification of twenty alternatives for preference-based abstract argumentation semantics using six principles suggests that our new reduction has some advantages over the existing ones, in the sense that if the set of preferences increases, the sets of accepted arguments increase as well. The results of this research have been published at COMMA [36].

Together with Celia da Costa Pereira (I3S) and Mauro Dragoni (FBK Trento), we presented SMACk, an opinion summary system built on top of an argumentation framework with the aim to exchange, communicate and resolve possibly conflicting viewpoints. SMACk allows the user to extract debated opinions from a set of documents containing user-generated content from online commercial websites, and to automatically identify the mostly debated positive aspects of the issue of the debate, as well as the mostly debated negative ones. The key advantage of such a framework is the combination of different methods, i.e., formal argumentation theory and natural language processing, to support users in making more informed decisions, e.g., in the context of online purchases. The results of this research have been published in the AI Communications journal [14].